Move serialization dirty into serialization.rb

This fixes a regression (#13744) that was caused by 67d8bb96.

In 67d8bb96, we introduced lazy rollback for records, such that the
record's internal states and attributes are not restored immediately
after a transaction rollback, but deferred until they are first
accessed.

This optimization is only performed when the model does not have any
transactional callbacks (e.g. `after_commit` and `after_create`).

Unfortunately, the models used to test the affected codepaths all
comes with some sort of transactional callbacks. Therefore this
codepath remains largely untested until now and as a result there are
a few issues in the implementation that remains hidden until now.

First, the `sync_with_transaction_state` (or more accurately,
`update_attributes_from_transaction_state`) would perform the
synchronization prematurely before a transaction is finalized (i.e.
comitted or rolled back). As a result, when the actuall rollback
happens, the record will incorrectly assumes that its internal states
match the transaction state, and neglect to perform the restore.

Second, `update_attributes_from_transaction_state` calls `committed!`
in some cases. This in turns checks for the `destroyed?` state which
also requires synchronization with the transaction stae, which causes
an infnite recurrsion.

This fix works by deferring the synchronization until the transaction
has been finalized (addressing the first point), and also unrolled
the `committed!` and `rolledback!` logic in-place (addressing the
second point).

It should be noted that the primary purpose of the `committed!` and
`rolledback!` methods are to trigger the relevant transactional
callbacks. Since this code path is only entered when there are no
transactional callbacks on the model, this shouldn't be necessary. By
unrolling the method calls, the intention here (to restore the states
when necessary) becomes more clear.

Currently Active Record can be configured via the environment variable `DATABASE_URL` or by manually injecting a hash of values which is what Rails does, reading in `database.yml` and setting Active Record appropriately. Active Record expects to be able to use `DATABASE_URL` without the use of Rails, and we cannot rip out this functionality without deprecating. This presents a problem though when both config is set, and a `DATABASE_URL` is present. Currently the `DATABASE_URL` should "win" and none of the values in `database.yml` are used. This is somewhat unexpected to me if I were to set values such as `pool` in the `production:` group of `database.yml` they are ignored.

There are many ways that active record initiates a connection today:

- Stand Alone (without rails)
  - `rake db:<tasks>`
  - ActiveRecord.establish_connection
 
- With Rails
  - `rake db:<tasks>`
  - `rails <server> | <console>`
  - `rails dbconsole`


We should make all of these behave exactly the same way. The best way to do this is to put all of this logic in one place so it is guaranteed to be used.

Here is my prosed matrix of how this behavior should work:

```
No database.yml
No DATABASE_URL
=> Error
```

```
database.yml present
No DATABASE_URL
=> Use database.yml configuration
```

```
No database.yml
DATABASE_URL present
=> use DATABASE_URL configuration
```

```
database.yml present
DATABASE_URL present
=> Merged into `url` sub key. If both specify `url` sub key, the `database.yml` `url`
   sub key "wins". If other paramaters `adapter` or `database` are specified in YAML,
   they are discarded as the `url` sub key "wins".
```

### Implementation

Current implementation uses `ActiveRecord::Base.configurations` to resolve and merge all connection information before returning. This is achieved through a utility class: `ActiveRecord::ConnectionHandling::MergeAndResolveDefaultUrlConfig`.

To understand the exact behavior of this class, it is best to review the behavior in activerecord/test/cases/connection_adapters/connection_handler_test.rb though it should match the above proposal.

* Move check from generated helper to test_help.rb, so that all
  applications can benefit
* Rather than just raising when the test schema has pending migrations,
  try to load in the schema and only raise if there are pending
  migrations afterwards
* Opt out of the check by setting
  config.active_record.maintain_test_schema = false
* Deprecate db:test:* tasks. The test helper is now fully responsible
  for maintaining the test schema, so we don't need rake tasks for this.
  This is also a speed improvement since we're no longer reloading the
  test database on every call to "rake test".

(No database connection) sounds not be
able to connect database.

see more #12804

Revert "ActiveRecord::Base#<=> has been removed.  Primary keys may not be in order," -- will be replaced with a check to ensure that the keys used for comparison are integers, and only fail if they are not.

This reverts commit 6256734e.

Conflicts:
	activerecord/CHANGELOG.md

Make AM::Dirty less dirty to plugin into AR or other library

This will make easier to hook protected_attributes gem in our code
without making that gem fragile to change in Rails code base.

Closes #12243

or even be numbers, so sorting by id doesn't make sense.  Please use `sort_by`
and specify the attribute you wish to sort with.  For example, change:

  Post.all.to_a.sort

to:

  Post.all.to_a.sort_by(&:id)

`disable_implicit_join_references=` was only deprecated on `master`,
not with rails 4.0. We can remove it after 4.1

This reverts commit 3c719ead, reversing
changes made to d5c3bf97.

Previously, freezing a cloned ActiveRecord object froze the original
too. By cloning `@attributes` before freezing, we prevent cloned objects
(which in Ruby share state of ivars) from being effected by `#freeze`.

Resolves issue #4936, which has further information on this issue, as
well as steps to reproduce.

* Add a test case for `#freeze` not causing `cloned.frozen?` to be true.
* Clone @attributes before freezing in `ActiveRecord::Core`, then
  reassign the cloned, frozen hash to the frozen model's `@attributes`
  ivar.

/cc @steveklabnik

Registries have class-level accessors to write clean code, let's
use them. This makes style uniform also with existing usage in
ScopeRegistry and InstrumentationRegistry.

If performance of the method_missing callback was ever considered to
be a concern, then we should stop using it altogether and probably
remove the callback. But while we have the feature we should use it.

When partial inserts are enabled, overridden db defaults are ignored. This
results in locking columns having a nil value for new records if the db default
is null. This happens because the list of changed attributes for new records is
always assumed to be empty.

Solution: When a new record's default attributes are set, also initialize the
list of changed attributes by comparing current values against what's stored as
the column defaults in the database.

Closes #9712.

This allows end-users to have a `connection` method on their models
without clashing with ActiveRecord internals.

Reduced memory leak problem in transactions by lazily updating AR objects with new transaction state. If AR object has a callback, the callback will be performed immediately (non-lazily) so the transaction still has to keep records with callbacks.

[ci skip]